yuzuki-natsumi / dsh-peak-avoidance

dsh-peak-avoidance

Auto-switch to an alternate model before DeepSeek peak hours and back to the official model afterwards — avoiding official peak-hour pricing.

Maintainer: 柚木夏实 · Code written by AI (DeepSeek Harness)

Disclaimer

This plugin was developed entirely by DSH (AI). The author has no programming experience, and the code has not been professionally reviewed. Users must test it themselves and judge its safety, stability, and correctness before use. The installation instructions in this repository may contain AI hallucinations (inaccurate or nonexistent steps, package names, or paths) — verify everything before executing. The author accepts no liability for any consequences of using this plugin.

Features

• Scheduled switching: enter avoidance at window start − leadMinutes (default 30), restore at window end.
• Both routes: agentDefaultModel (default route) + apiProxy.sessions.selectModel (live sessions, takes effect next turn).
• Multiple windows, cross-midnight supported (e.g. 22:00-02:00).
• Reasoning effort per model (high, …), dropdown populated from the live model catalog.
• Provider/model dropdowns fetched live from the llm service.
• Switch toast: frame-wide banner on enter/exit (toggleable; polling stops when off).
• Manual control: card buttons plus model tools peak_avoidance_status / peak_avoidance_control.
• Safe: sessions manually reverted during peak are never forced back; originals are restored on stop (restoreOnStop).

Install

Option 1 — npm (recommended for DSH users):

npm install dsh-peak-avoidance

The @deepseek-ai runtime deps are peerDependencies, provided by DSH itself — nothing extra to install. Installing outside DSH requires providing the peers (@deepseek-ai/dsh-settings, @deepseek-ai/dsh-tools, @deepseek-ai/cordis) yourself.

Option 2 — from source: clone this repo and place lib/ + package.json into a DSH-resolvable directory (e.g. $DSH_HOME/profiles/node_modules/dsh-peak-avoidance/).

Both options then require appending to $DSH_HOME/cordis.patch.yml:

- insert:
    - id: peak-avoidance
      name: 'dsh-peak-avoidance'

and restarting DSH.

Usage

• Settings → Plugins → Configurable → Peak Avoidance card: status, config, manual enter/restore.
• Config lives in DSH settings (peak-avoidance section of settings.yaml); changes apply immediately.
• Required: peakModel (avoidance target) and normalModel (official model; copy the current default on first run).

Privacy & security

• No API keys are read, stored, or emitted; only model selections and its own config.
• The client bridge accepts loopback requests only.
• All catalog/model/effort data comes live from the DSH llm service — no deployment-specific hardcoding.

Development

• No build step: lib/index.js (host, Cordis plugin module) + lib/client.js (browser, module-loader format).
• Deps: @deepseek-ai/dsh-settings, @deepseek-ai/dsh-tools, @deepseek-ai/schemastery.

License

MIT